TDeint and TIVTC - Page 56

tritical · 13th January 2008, 01:54

It's caused by the combination of tfm's internal motion masking (PP>4) and the external deinterlacing. If you set PP=4 instead of PP=7 (or use only nnedi as the clip2 input) the problem goes away. It's a good idea to disable tfm's internal motion masking, which is run post field matching, when using TDeint (or another motion-adaptive deinterlacer) for post-processing since TDeint's/TMM's motion masking is run on the original stream.

Fizick · 14th January 2008, 20:30

Recently I saw some thread about comparing various deinterlacers (and its modes) by speed and quality. Some mode of TDeint was a winner (optimal). I want point some user to it, but I can not find it. Anybody can pont me to this thread? Or it was deleted?

jeffy · 15th January 2008, 00:23

Quote:

Originally Posted by Fizick

Recently I saw some thread about comparing various deinterlacers (and its modes) by speed and quality. Some mode of TDeint was a winner (optimal). I want point some user to it, but I can not find it. Anybody can pont me to this thread? Or it was deleted?

Some hints:
http://forum.doom9.org/showthread.ph...16#post1059016

http://heptium.sh.cvut.cz/~integra/deint/
linked from:
http://forum.doom9.org/showthread.ph...011#post900011

Chainmax · 15th January 2008, 02:43

Quote:

Originally Posted by tritical

It's caused by the combination of tfm's internal motion masking (PP>4) and the external deinterlacing. If you set PP=4 instead of PP=7 (or use only nnedi as the clip2 input) the problem goes away. It's a good idea to disable tfm's internal motion masking, which is run post field matching, when using TDeint (or another motion-adaptive deinterlacer) for post-processing since TDeint's/TMM's motion masking is run on the original stream.

Are you sure? Like I said in the post with the pictures, only modes 3 and 5 caused it, all other modes worked fine.

tritical · 15th January 2008, 09:40

@Chainmax
Well, I was sure until I started to write an explanation of why it occurs, and it didn't actually make sense. Turns out there was a bug in modes 3/5 with micmatching > 0, which would cause tfm to sometimes return a u or b match when it thought it was returning a p, c, or n match. That error could only happen if all 5 matches were combed. I'll put up a fixed version later tonight or tommorrow.

@Fizick
I can't recall reading a thread comparing deinterlacers by both speed and quality.

EDIT:

Maybe it was this thread: http://forum.doom9.org/showthread.php?t=117025 ? It has some speed/filesize numbers plus some quality (ssim/psnr) numbers.

I did a comparison back in April 07 that I never finished, but I think that when taking both speed and quality (ssim/psnr wise) into account that it is difficult to beat:

yadifmod(mode=1,edeint=tdeint(mthreshL=0,mthreshC=0))

If only quality is considered than mcbob/mvbob are the clear winners (usually mcbob is best). If only speed is considered than the winners are leakkerneldeint/bob, yadif, or tomsmocomp (leaving out avisynth's internal bob() of course).

Fizick · 15th January 2008, 18:07

Thanks for responds, but it was some other thread...
I found it !

http://forum.doom9.org/showthread.php?t=133078

tritical · 17th January 2008, 08:25

TIVTC v1.0.5, changes:

Code:

- Fixed a bug that would cause tfm to return a 'u' or 'b' match when it thought it was
     returning a 'p', 'c', or 'n' match.  This problem only occured if mode was set to
     3 or 5, micmatching was enabled (not set to 0), and all five matches were combed.

Mottodj · 1st February 2008, 23:33

Hi everybody! I have question about results field maching by TFM. I have DVD with NTSC(film but without pulldown flags) anime and i pass it through TFM using avisynth. Field maching is nice but there is artifact in picture that was already mached(on fade to dark and in other way). Lines on picture have different brightness and it changing each other(ex. http://mini.100-games.ru/4super/artifact.png ). Have you ever seen that thing, what to do with this? Passing through tomsmocomp don't produce such artifact but tomsmocomp blur picture and it isn't solution...

Dreassica · 2nd February 2008, 00:26

That's not missed interlace, but common in fades. I usually just applyrange(x,x,"blur",0,0.7) it, works perfectly.

Leak · 3rd February 2008, 04:13

Quote:

Originally Posted by Dreassica

That's not missed interlace, but common in fades.

To be totally precise, that's a fade that has been applied AFTER the telecine was applied, so you have a 30 frames (or maybe even 60 fields) per second fade over pulled down 24 FPS material, so you can have either of them cleaned up, but not both - but then again, in a fade it's usually not noticeable UNLESS of course you take a screenshot...

A similar thing would be 30p/60i end credits (for extra smoothness that's easily produced) scrolling over 24p animation, which also happens often.

foxyshadis · 3rd February 2008, 12:58

Use Vinverse for mild residual combing, not a destructive blur. Works wonders with some post-telecine editing, unless it's a very fast fade.

NorthPole · 5th February 2008, 05:06

I've been using the following sort-of generic catch-all script for re-encoding ATSC digital broadcast caps to 23.976. I had a couple of questions..

Do I need to determine the order or does DGDecode_mpeg2source set the order correctly?

When converting from 59.94 fps progressive to 23.976 fps, sometimes I get some jerkiness in using the tdecimate(cycler=3, cycle=5). Is there anything I could do to tweak this?

I'm kind of lazy about counting frames in VirtualDub....Sorry

I have used the tdecimate(hybrid=1) setting on 29.97 fps to 23.976 fps and seem to get good results but I don't think I can use that for 59.94 fps conversions.

Any input would be appreciated.

PS.. Thanks tritical for you work!

Code:

DGDecode_mpeg2source("__vid__",info=3)
LoadPlugin("C:\Program Files\Media\AviSynth\plugins\tivtc.dll")
LoadPlugin("C:\Program Files\Media\AviSynth\plugins\eedi2.dll")
LoadPlugin("C:\Program Files\Media\AviSynth\plugins\tdeint.dll")
(last.height == 480)? SD(last): HD(last)
last.letterbox(4,2,2,2) # to mask screen edge noise

Function SD(video) {
ord = video.getparity() ? 1 : 0
ord == 1 ? video.AssumeTFF() : video.AssumeBFF()
edeintted = video.SeparateFields().SelectEven().EEDI2(field=ord)
tdeintted = video.TDeint(edeint=edeintted,order=ord,hints=false)
video.tfm(order=ord,clip2=tdeintted).tdecimate(hybrid=1)
#last.Lanczos4Resize(720,480,0,0,704,480)
}

Function HD(video) {
(video.height == 720)?
\ video.tfm().tdecimate(cycler=3,cycle=5):
\ video.tfm().tdecimate(hybrid=1)
}

foxyshadis · 5th February 2008, 05:47

Quote:

Originally Posted by NorthPole

Do I need to determine the order or does DGDecode_mpeg2source set the order correctly?

mpeg2source will do that.

Quote:

When converting from 59.94 fps progressive to 23.976 fps, sometimes I get some jerkiness in using the tdecimate(cycler=3, cycle=5). Is there anything I could do to tweak this?

Well, not everything is telecined. You're sure these are, rather than true 30 or even 60 fps segments? You can try tweaking chroma, nt, and dupthresh parameters to see if they help your stuff.

Quote:

I have used the tdecimate(hybrid=1) setting on 29.97 fps to 23.976 fps and seem to get good results but I don't think I can use that for 59.94 fps conversions.

You'll get better results out of mvflowfps if you must force an fps change.

NorthPole · 5th February 2008, 15:28

@foxyshadis
Thank you for the input.

Quote:

Originally Posted by foxyshadis

mpeg2source will do that.

Good, I'll take it out.

Quote:

Originally Posted by foxyshadis

Well, not everything is telecined. You're sure these are, rather than true 30 or even 60 fps segments? You can try tweaking chroma, nt, and dupthresh parameters to see if they help your stuff.

I believe most, if not all, is just 24p slowed down and telecined... Mostly movies and sitcoms.

If I understand the hybrid mode, it does normal 1 out of 5 decimation for 30 fps video segments provided it has a match. If it doesn't then it does blended decimation to get to 24 fps. I was kind of looking for something like that for 59.94 fps or 60 fps.

like say this...
tfm().tdecimate(cycle=2).tdecimate(hybrid=1)

I will try the dupThresh, chroma and nt options but then I think I have to use mode 1. So perhaps say something like..

tfm().tdecimate(mode=1, dupThresh=0.9, chroma=true, nt=1, cycler=3,cycle=5)
or
tfm().tdecimate(mode=1, dupThresh=1.2, chroma=false, nt=1, cycler=3,cycle=5)

Quote:

Originally Posted by foxyshadis

You'll get better results out of mvflowfps if you must force an fps change.

I'm not sure what this is?

NorthPole · 5th February 2008, 22:30

I did a couple of test encodes on a 1280 x 720p ATSC cap with:

tdecimate(cycler=3,cycle=5)
and
tdecimate(cycle=2).tdecimate(hybrid=1)

I wasn't thinking earlier but I wouldn't need the field matcher, tfm, since the source is all progressive.

the frames pattern go as follows: 1 0 0 1 0 1 0 0 1 0 1 0 0 1 0 etc.

1 = new frame, 0 = duplicate frame

I get the correct resulting frame selections with both methods and the encoding time is very similar. I believe that tdecimate(cycle=2).tdecimate(hybrid=1) might be a better way to go because it could correctly handle true 24 fps and 30 fps sources. The 24 fps source would be returned to the original frame rate and the 30 fps source would be blend decimated to 24 fps and hopefully avoid any jerkiness.

I tried the dupThresh, chroma and nt options with no noticably difference however, unfortunately, I deleted the original 720p video clip that was giving me jerkiness. I will have more in the next week or two and will report whether those tweaks helped.

rebkell · 5th February 2008, 23:07

ABC is notorious for dropping frames, one of their favorite sequences is:
3:2
3:2
3:2
3:2
3:2
3:2
3:2
3:2
3:1
3:2
3:2
3:2
3:2
3:2
3:2
3:2
2:2

and then they repeat, it will cause the jerkiness when trying to get to the 23.9675 FPS. It works out to actually be 24.553735 FPS. I've seen it on several shows, LOST, Men In Trees, and some others, occasionally they will add an extra frame.

Adub · 6th February 2008, 02:34

Ok, so how do you deal with it?

rebkell · 6th February 2008, 03:31

Quote:

Originally Posted by Merlin7777

Ok, so how do you deal with it?

Check out this thread, I'm not sure why it's in the DVD2AVI / DGIndex section, but it is.

http://forum.doom9.org/showthread.php?t=122472

I use FDecimate, I use megui and I actually built a profile to step through the frames, normally you can determine the cadence, basically you use the metrics=true to try to determine the best threshold value to use and of course after stepping through you can determine the FPS. The cadence I posted is the most frequent one I've run into, it's not perfect, but if you can figure out a good threshold value it works pretty well. All the above will make more sense when you read the thread I linked to above. I've also run into 10 consecutive 3:2 sequences and 1 2:2, that repeats every 11 cadences(I guess that's the correct word)

I've seen some cadences on the Fox Network that I never could determine any pattern for.

Adub · 6th February 2008, 06:44

Thank you for that. I have to learn how to count fields. Anyone have a "guide" for doing it?

NorthPole · 6th February 2008, 15:23

@rebkill,

Thanks for your input... I haven't counted as many frames as you have (not fun). I will check out the thread you suggested.

@merlin777,

I just count frames in virtualdub as I go through the video manually. I don't know of any specific method or program that does it automatically.

15th January 2008, 09:40	#1105 \| Link
tritical Registered User Join Date: Dec 2003 Location: MO, US Posts: 999	@Chainmax Well, I was sure until I started to write an explanation of why it occurs, and it didn't actually make sense. Turns out there was a bug in modes 3/5 with micmatching > 0, which would cause tfm to sometimes return a u or b match when it thought it was returning a p, c, or n match. That error could only happen if all 5 matches were combed. I'll put up a fixed version later tonight or tommorrow. @Fizick I can't recall reading a thread comparing deinterlacers by both speed and quality. EDIT: Maybe it was this thread: http://forum.doom9.org/showthread.php?t=117025 ? It has some speed/filesize numbers plus some quality (ssim/psnr) numbers. I did a comparison back in April 07 that I never finished, but I think that when taking both speed and quality (ssim/psnr wise) into account that it is difficult to beat: yadifmod(mode=1,edeint=tdeint(mthreshL=0,mthreshC=0)) If only quality is considered than mcbob/mvbob are the clear winners (usually mcbob is best). If only speed is considered than the winners are leakkerneldeint/bob, yadif, or tomsmocomp (leaving out avisynth's internal bob() of course). Last edited by tritical; 15th January 2008 at 10:21.

5th February 2008, 05:06	#1112 \| Link
NorthPole Registered User Join Date: Jan 2006 Posts: 141	I've been using the following sort-of generic catch-all script for re-encoding ATSC digital broadcast caps to 23.976. I had a couple of questions.. Do I need to determine the order or does DGDecode_mpeg2source set the order correctly? When converting from 59.94 fps progressive to 23.976 fps, sometimes I get some jerkiness in using the tdecimate(cycler=3, cycle=5). Is there anything I could do to tweak this? I'm kind of lazy about counting frames in VirtualDub....Sorry I have used the tdecimate(hybrid=1) setting on 29.97 fps to 23.976 fps and seem to get good results but I don't think I can use that for 59.94 fps conversions. Any input would be appreciated. PS.. Thanks tritical for you work! Code: DGDecode_mpeg2source("__vid__",info=3) LoadPlugin("C:\Program Files\Media\AviSynth\plugins\tivtc.dll") LoadPlugin("C:\Program Files\Media\AviSynth\plugins\eedi2.dll") LoadPlugin("C:\Program Files\Media\AviSynth\plugins\tdeint.dll") (last.height == 480)? SD(last): HD(last) last.letterbox(4,2,2,2) # to mask screen edge noise Function SD(video) { ord = video.getparity() ? 1 : 0 ord == 1 ? video.AssumeTFF() : video.AssumeBFF() edeintted = video.SeparateFields().SelectEven().EEDI2(field=ord) tdeintted = video.TDeint(edeint=edeintted,order=ord,hints=false) video.tfm(order=ord,clip2=tdeintted).tdecimate(hybrid=1) #last.Lanczos4Resize(720,480,0,0,704,480) } Function HD(video) { (video.height == 720)? \ video.tfm().tdecimate(cycler=3,cycle=5): \ video.tfm().tdecimate(hybrid=1) }

6th February 2008, 02:34	#1117 \| Link
Adub Fighting spam with a fish Join Date: Sep 2005 Posts: 2,699	Ok, so how do you deal with it? __________________ FAQs:Bond's AVC/H.264 FAQ Site:Adubvideo

6th February 2008, 06:44	#1119 \| Link
Adub Fighting spam with a fish Join Date: Sep 2005 Posts: 2,699	Thank you for that. I have to learn how to count fields. Anyone have a "guide" for doing it? __________________ FAQs:Bond's AVC/H.264 FAQ Site:Adubvideo

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

13th January 2008, 01:54	#1101 \| Link
tritical Registered User Join Date: Dec 2003 Location: MO, US Posts: 999	It's caused by the combination of tfm's internal motion masking (PP>4) and the external deinterlacing. If you set PP=4 instead of PP=7 (or use only nnedi as the clip2 input) the problem goes away. It's a good idea to disable tfm's internal motion masking, which is run post field matching, when using TDeint (or another motion-adaptive deinterlacer) for post-processing since TDeint's/TMM's motion masking is run on the original stream.

14th January 2008, 20:30	#1102 \| Link
Fizick AviSynth plugger Join Date: Nov 2003 Location: Russia Posts: 2,183	Recently I saw some thread about comparing various deinterlacers (and its modes) by speed and quality. Some mode of TDeint was a winner (optimal). I want point some user to it, but I can not find it. Anybody can pont me to this thread? Or it was deleted?

15th January 2008, 18:07	#1106 \| Link
Fizick AviSynth plugger Join Date: Nov 2003 Location: Russia Posts: 2,183	Thanks for responds, but it was some other thread... I found it ! http://forum.doom9.org/showthread.php?t=133078

1st February 2008, 23:33	#1108 \| Link
Mottodj Registered User Join Date: Jan 2008 Posts: 5	Hi everybody! I have question about results field maching by TFM. I have DVD with NTSC(film but without pulldown flags) anime and i pass it through TFM using avisynth. Field maching is nice but there is artifact in picture that was already mached(on fade to dark and in other way). Lines on picture have different brightness and it changing each other(ex. http://mini.100-games.ru/4super/artifact.png ). Have you ever seen that thing, what to do with this? Passing through tomsmocomp don't produce such artifact but tomsmocomp blur picture and it isn't solution...

2nd February 2008, 00:26	#1109 \| Link
Dreassica Registered User Join Date: May 2002 Posts: 384	That's not missed interlace, but common in fades. I usually just applyrange(x,x,"blur",0,0.7) it, works perfectly.

3rd February 2008, 12:58	#1111 \| Link
foxyshadis ангел смерти Join Date: Nov 2004 Location: Lost Posts: 9,558	Use Vinverse for mild residual combing, not a destructive blur. Works wonders with some post-telecine editing, unless it's a very fast fade.

5th February 2008, 22:30	#1115 \| Link
NorthPole Registered User Join Date: Jan 2006 Posts: 141	I did a couple of test encodes on a 1280 x 720p ATSC cap with: tdecimate(cycler=3,cycle=5) and tdecimate(cycle=2).tdecimate(hybrid=1) I wasn't thinking earlier but I wouldn't need the field matcher, tfm, since the source is all progressive. the frames pattern go as follows: 1 0 0 1 0 1 0 0 1 0 1 0 0 1 0 etc. 1 = new frame, 0 = duplicate frame I get the correct resulting frame selections with both methods and the encoding time is very similar. I believe that tdecimate(cycle=2).tdecimate(hybrid=1) might be a better way to go because it could correctly handle true 24 fps and 30 fps sources. The 24 fps source would be returned to the original frame rate and the 30 fps source would be blend decimated to 24 fps and hopefully avoid any jerkiness. I tried the dupThresh, chroma and nt options with no noticably difference however, unfortunately, I deleted the original 720p video clip that was giving me jerkiness. I will have more in the next week or two and will report whether those tweaks helped.

5th February 2008, 23:07	#1116 \| Link
rebkell Registered User Join Date: Oct 2006 Posts: 303	ABC is notorious for dropping frames, one of their favorite sequences is: 3:2 3:2 3:2 3:2 3:2 3:2 3:2 3:2 3:1 3:2 3:2 3:2 3:2 3:2 3:2 3:2 2:2 and then they repeat, it will cause the jerkiness when trying to get to the 23.9675 FPS. It works out to actually be 24.553735 FPS. I've seen it on several shows, LOST, Men In Trees, and some others, occasionally they will add an extra frame.

6th February 2008, 15:23	#1120 \| Link
NorthPole Registered User Join Date: Jan 2006 Posts: 141	@rebkill, Thanks for your input... I haven't counted as many frames as you have (not fun). I will check out the thread you suggested. @merlin777, I just count frames in virtualdub as I go through the video manually. I don't know of any specific method or program that does it automatically.